Table 10 Summary of the advantages of signgaurd over existing techniques.
From: A hybrid machine learning framework for offline signature verification using gray wolf optimization
Ref. | Limitation of Existing Work | Advantages of SignGaurd |
|---|---|---|
LBP designed for general texture classification, not optimized for signatures. | Introduces CS-LBP and OC-CSLBP, tailored for offline signature features, improving accuracy to 98.77%. | |
Low accuracy (~ 67%) for Indic scripts using LBP/ULBP. | OC-CSLBP capture both local & global textures efficiently. | |
LDP effective only for binary/black-and-white signatures, not generalized. | Hybrid features (CS-LBP + OC-CSLBP) work on grayscale signatures, ensuring better generalization. | |
One-Class SVM sensitive to distance metrics and threshold settings. | Hybrid ML framework (SVM + XGBoost) reduces overfitting, stabilizes threshold sensitivity, and lowers FAR/FRR. | |
High error rate (15.41%) for skilled forgeries. | GWO preprocessing + OC-CSLBP provide discriminative features, reducing FAR to 0.38% and FRR to 0%. | |
Only marginal AER reduction, limited improvement. | OC-CSLBP halves feature size and increases accuracy, that ensures robustness. | |
Very low FRR (2%) but high FAR (11%), less security. | Hybrid model achieves balanced FAR (0.38%) and FRR (0%), improving reliability. | |
Triangular geometric features gave very high AER (34%). | Texture-based OC-CSLBP features outperform geometric-only features with much lower error rates. | |
ANN with simple geometric features gave only moderate accuracy (86.67%). | Advanced handcrafted descriptors + SVM-XGBoost improves accuracy to more than 99%. | |
Hierarchical clustering achieved only 80% accuracy on small dataset. | Writer-independent framework generalizes better, validated with CEDAR dataset, yielding 99%+. | |
CNNs achieved high accuracy but required large datasets & high computation. | SignGuard achieves comparable accuracy using lightweight features & hybrid ML, suitable for small datasets and low-resource systems. | |
Focused on writer identification, not verification. | Verification-specific framework ensures direct applicability to signature authentication. | |
High accuracy but poor generalization across datasets. | POA + GWO preprocessing improve robustness across variations. | |
Type-2 neutrosophic logic adds computational complexity. | OC-CSLBP reduces feature vector size by 50%, making the model more efficient. | |
One-class learning less effective when forged samples are available. | SignGuard trains with both genuine and forged signatures, improving detection of skilled forgeries. |
